Statistical region-based segmentation methods such as the Active Appearance Models establish dense correspondences by modelling variation of shape and pixel intensities in low-resolution 2D images. Unfortunately, for high-resolution 2D and 3D images, this approach is rendered infeasible due to excessive storage and computational requirements. This paper addresses the problem by modelling the appearance of wavelet coefficient subsets contrary to the pixel intensities. We call this Wavelet Enhanced Appearance Modelling (WHAM). Experiments using the orthogonal Haar wavelet and the bi-orthogonal CDF 9-7 wavelet on cardiac MRIs and human faces show that the segmentation accuracy is minimally degraded at compression ratios of 1:10 and 1:20, respectively.
|Title of host publication||Proc. 11th Danish Conference on Pattern Recognition and Image Analysis|
|Publisher||Department of Computer Science, University of Copenhagen (DIKU)|
|Publication status||Published - 2002|
Stegmann, M. B., Forchhammer, S., & Olsen, S. I. (Ed.) (2002). On exploiting wavelet bases in statistical region-based segmentation. In Proc. 11th Danish Conference on Pattern Recognition and Image Analysis (pp. 75-82). Department of Computer Science, University of Copenhagen (DIKU). http://www.imm.dtu.dk/pubdb/p.php?857